American Journal of Epidemiology ª The Author 2009. Published by the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail:
[email protected].
Vol. 169, No. 9 DOI: 10.1093/aje/kwp021 Advance Access publication March 18, 2009
Original Contribution Is House-Dust Nicotine a Good Surrogate for Household Smoking?
Todd Whitehead, Catherine Metayer, Mary H. Ward, Marcia G. Nishioka, Robert Gunier, Joanne S. Colt, Peggy Reynolds, Steve Selvin, Patricia Buffler, and Stephen M. Rappaport Initially submitted August 13, 2008; accepted for publication January 15, 2009.
The literature is inconsistent regarding associations between parental smoking and childhood leukemia, possibly because previous studies used self-reported smoking habits as surrogates for children’s true exposures to cigarette smoke. Here, the authors investigated the use of nicotine concentrations in house dust as measures of children’s exposure to cigarette smoke in 469 households from the Northern California Childhood Leukemia Study (1999–2007). House dust was collected by using high-volume surface samplers and household vacuum cleaners and was analyzed for nicotine via gas chromatography–mass spectrometry. Using multivariable linear regression, the authors evaluated the effects of self-reported parental smoking, parental demographics, house characteristics, and other covariates on house-dust nicotine concentrations. They observed that nicotine concentrations in house dust were associated with self-reported smoking for periods of months and years before dust collection. Furthermore, the authors found that the relation between nicotine dust levels and self-reported smoking varied by parental age and socioeconomic status. These findings suggest that house-dust nicotine concentrations reflect long-term exposures to cigarette smoke in the home and that they may be less biased surrogates for children’s exposures to cigarette smoke than self-reported smoking habits. child; dust; environmental exposure; infant; leukemia; linear models; nicotine; smoking
Abbreviations: HVS3, high-volume surface sampler; KW ANOVA, Kruskal-Wallis one-way analysis of variance; NCCLS, Northern California Childhood Leukemia Study; SES, socioeconomic status.
Although parental smoking is a potential contributor to childhood leukemia risk, the evidence supporting such an association is inconsistent. Studies have suggested variously that maternal smoking during pregnancy increases the risk of acute lymphocytic leukemia (1–3) and acute myeloid leukemia (4, 5); that paternal smoking increases the risk of infant leukemia (6), acute lymphocytic leukemia (7–10), and acute myeloid leukemia (8, 10); and that neither maternal nor paternal smoking is associated with childhood leukemia (11–15). All previous research on potential associations between childhood leukemia and parental smoking has relied on selfreported smoking histories, which can result in misclassification of children’s true exposures to cigarette smoke and subsequent bias in the exposure-response relation (16). Studies using nicotine-specific cotinine biomarkers as ‘‘gold’’ standards have shown that only about 5% of pro-
fessed nonsmokers are actually smokers (17–19); however, deception rates as high as 25% have been observed when parents report their smoking habits in studies involving their children’s health (20, 21). Interestingly, researchers observed that for 11,083 self-reported nonsmokers, the likelihood of an elevated urinary cotinine level (consistent with smoking) decreased with increasing education (22). This finding indicates that less educated subjects may be more likely to underreport their smoking exposure. To reduce misclassification of exposure to cigarette smoke, it is beneficial to use an objective measure of exposure such as nicotine in indoor air (23), cotinine in urine (24), or nicotine in hair (25). Alternatively, researchers have suggested using nicotine levels in house dust as surrogates for in-home exposures to cigarette smoke (26–29). Indeed, previous research has shown that nicotine concentrations in
Correspondence to Todd Whitehead, School of Public Health, University of California, 50 University Hall, MC 7360, Berkeley, CA 94720 (e-mail:
[email protected]).
1113
Am J Epidemiol 2009;169:1113–1123
1114 Whitehead et al.
house dust are highly correlated with children’s levels of urinary cotinine (rS ¼ 0.77, n ¼ 15) in households with smokers (28). However, previous investigations of nicotine levels in house dust (26–29) involved small numbers of households (n ¼ 72, n ¼ 49, n ¼ 23, and n ¼ 37, respectively) and were unable to thoroughly examine the determinants of house-dust nicotine concentrations. In this paper, we report results from 469 households in which nicotine was measured in house dust and from which extensive questionnaire data, including smoking habits, were also obtained from residents. These data were collected as part of the Northern California Childhood Leukemia Study (NCCLS), a large case-control study (8). The objectives of the current study were to compare house-dust nicotine levels with self-reported smoking at various times before and during a child’s life and to identify determinants of house-dust nicotine levels. Although this information is directly relevant to researchers considering the effect of parental smoking on childhood leukemia risk, it should also be pertinent for any epidemiologic study that seeks to quantify exposure to cigarette smoke. MATERIALS AND METHODS Study population
The NCCLS is an ongoing study conducted in the San Francisco Bay Area and California Central Valley in which cases aged 0–14 years are ascertained from 9 pediatric clinical centers. Controls, matched to cases on date of birth, gender, race, Hispanic ethnicity, and maternal residence, are selected from the California birth registry (8). The homes of cases and controls aged 0–7 years who lived at the home they occupied at the time of diagnosis (and a similar reference date for controls) from December 1999 through November 2007 were eligible for household dust collection. Among 324 cases and 407 controls determined to be eligible, 296 cases (91%) and 333 controls (82%) participated. We obtained informed consent from the children’s parent or legal guardian in accordance with the institutional review boards’ requirements at the University of California, Berkeley; the National Cancer Institute; and other participating institutions. House-dust nicotine measurement
Dust was collected by using a high-volume surface sampler (HVS3) in the room where the child spent the most time while awake (commonly the family room) as well as from household vacuum cleaners, as previously described (30); data derived from both methods were used in our analyses. In HVS3-sampled homes (82%), we recorded the area of the carpet sampled, but this information was not relevant in homes where vacuum cleaner dust was used. For the nicotine analyses, each 0.5-g dust aliquot was spiked with 250 ng of d4-nicotine, extracted by ultrasonication in dichloromethane, concentrated, and analyzed by using a gas chromatograph–mass spectrometer in the multiple ion detection mode. The gas chromatograph analysis utilized a DB-1701 column (30 m, 0.25-mm internal diameter, 0.15-lm film) that was programmed from 130C to 220C at
2C/minute and then from 220C to 280C at 10C/minute. Dibromobiphenyl was used as an internal standard; a 9-point calibration curve (range: 2–750 ng/mL) and a zerolevel standard were analyzed with each sample set (12 field samples, a duplicate, a duplicate spike (250 ng), and a solvent method blank). We used d4-nicotine as a surrogate recovery standard to correct for variable nicotine recovery on a sample-by-sample basis. Recoveries of nicotine and d4nicotine in spiked samples were 57% (standard deviation, 45) and 59% (standard deviation, 45), respectively. The median relative percentage difference for duplicates was 17% after surrogate recovery standard correction. Self-reported smoking
Parents, primarily the mother (97%), responded to 2 sets of questionnaires, each with inquiries about smoking habits, as outlined in Figure 1. The initial interview ascertained the smoking status of the mother, father, and others in the household at several time points of interest (Table 1). Additionally, the first interview asked the respondent for the number of cigarettes smoked per day for some but not all of the time periods. A subsequent interview at the time of dust collection ascertained the total number of cigarettes smoked per day inside the house during the previous month. This additional question dealt specifically with smoking inside the home and was therefore expected to correspond to the concurrent house-dust nicotine measurements. However, we also considered responses from the first interview as potential determinants of house-dust nicotine concentrations because nicotine is known to persist indoors (31), where it is protected from degradation by moisture, sunlight, and microbial action. Statistical analysis
Because the distribution of nicotine in house dust was highly skewed, the nonparametric Kruskal-Wallis one-way analysis of variance (KW ANOVA) was used to compare the distribution of nicotine in house dust between various groups throughout the analysis. The data had an approximate lognormal distribution, so the natural log of the concentration was used in all analyses involving the continuous variable. House-dust nicotine measurements below the limit of detection, that is, 20 ng/g (n ¼ 53, 11% of households), were assigned a value of one-half the limit of detection. Pairwise correlation coefficients between the natural log-transformed house-dust nicotine concentrations and self-reported cigarette smoking (as well as other variables of interest) were estimated. Although Pearson correlation coefficients are reported here, results were similar when we used Spearman rank coefficients. Seven groups of variables were considered for inclusion in the house-dust nicotine regression models: self-reported smoking, parental demographics, house characteristics, child-specific variables, sampling conditions, time effects, and ethnicity (refer to Appendix Table 1 for the full list of variables considered). Groups of highly correlated variables were examined by principal components analysis to produce simpler, but meaningful, summary measures of the variables within these groups for inclusion in the final house-dust nicotine regression models (32). The remaining groups of Am J Epidemiol 2009;169:1113–1123
House-Dust Nicotine
1115
Figure 1. Conceptual timeline of the Northern California Childhood Leukemia Study, 1999–2007, showing time variables included in the statistical analysis as potential modifiers of house-dust nicotine concentrations. For each variable, the median value for the study population (measured in years) is shown in parentheses.
candidate variables were modeled individually by using backward elimination (P < 0.10) to identify other variables used in the final models. In addition to main effects, we included significant interactions (P < 0.10) between self-reported smoking variables and parental demographic variables and between selfreported smoking variables and case-control status. Using the variables identified in group screening, we performed 3 subsequent regression analyses with case and control households combined. The first analysis used all possible households regardless of the sampling method (both HVS3 and vacuum cleaner dust samples were used). The second analysis used only HVS3-sampled households; this analysis included size of sampling area, a variable relevant to only those homes where HVS3 sampling was conducted. The third analysis evaluated the effect of data censoring (due to values below the limit of detection) on the regression coefficients from the
first model. We used Tobit regression to model the logged house-dust nicotine concentrations, estimating the parameters of the uncensored data by using maximum likelihood and assuming a normally distributed error term. RESULTS Nicotine in house dust
Our analysis included 233 cases and 236 controls for whom house-dust nicotine measurements were available. Nicotine was detected in 89% (416 of 469) of the households. The nicotine concentrations ranged from not detected (less than 20 ng/g) to a maximum of 35,000 ng/g, with a median value of 265 ng/g and an interquartile range of 96–612 ng/g. Table 1 shows the prevalence of smoking
Table 1. Prevalence of Smoking at Various Time Periods Indicated by Variables Derived From Interviews, Northern California Childhood Leukemia Study, 1999–2007 Median Nicotine Concentration (ng/g)
Response Self-reported Cigarette Smoking Variable Yes
Mother ever smoked
No
% Yes
Yes
No
P Valuea
